Re: [qemu-s390x] [PATCH v3 8/9] target/s390x: Skip wout, cout helpers if op helper does not return

[Richard Henderson]

On 10/11/18 1:06 AM, David Hildenbrand wrote:
> On 03/10/2018 21:39, Richard Henderson wrote:
>> When op raises an exception, it may not have initialized the output
>> temps that would be written back by wout or cout.
>>
>> Cc: address@hidden
>> Signed-off-by: Richard Henderson <address@hidden>
>> ---
>>  target/s390x/translate.c | 12 +++++++-----
>>  1 file changed, 7 insertions(+), 5 deletions(-)
>>
>> diff --git a/target/s390x/translate.c b/target/s390x/translate.c
>> index 7363aabf3a..7fad3ad8e9 100644
>> --- a/target/s390x/translate.c
>> +++ b/target/s390x/translate.c
>> @@ -6164,11 +6164,13 @@ static DisasJumpType translate_one(CPUS390XState 
>> *env, DisasContext *s)
>>      if (insn->help_op) {
>>          ret = insn->help_op(s, &o);
>>      }
>> -    if (insn->help_wout) {
>> -        insn->help_wout(s, &f, &o);
>> -    }
>> -    if (insn->help_cout) {
>> -        insn->help_cout(s, &o);
>> +    if (ret != DISAS_NORETURN) {
>> +        if (insn->help_wout) {
>> +            insn->help_wout(s, &f, &o);
>> +        }
>> +        if (insn->help_cout) {
>> +            insn->help_cout(s, &o);
>> +        }
>>      }
>>  
>>      /* Free any temporaries created by the helpers.  */
>>
> 
> What about things like LPSW/LPWSE ? They certainly don't imply that we
> had an exception.

Exception in the tcg sense, not the guest architectural sense, in that we call
cpu_loop_exit from the helper, which performs a longjmp.  (Incidentally,
there's no reason to do that for load_psw -- we could just exit the tb 
normally.)

> (these two don't use wout/cout, so it is still fine, but I would prefer
> a comment somewhere because otherwise it is really easy to miss that
> DISAS_NORETURN makes us skip these handlers)

Where would you like me to place that comment?  In the DisasInsn definition?


r~